The Loria Instruction Generation System L in GIVE 2.5

Alexandre A. J. Denis (1)

1 TALARIS - Natural Language Processing: representation, inference and semantics

This paper presents the instruction generation system L submitted by the LORIA and TALARIS team to the GIVE challenge 2011 (GIVE 2.5). The system L takes the same approach to instruction generation than its predecessor the system NA that participated to the GIVE challenge 2010 (GIVE 2), the two systems are almost the same except minor modifications. We present the strategy of these systems, namely a directive, low level, navigation strategy ("Go left") and a referring strategy based on focus and sub-contexts (Denis, 2010) ("Not this one! Look for the other one"). These strategies were successful, as shown by the GIVE 2 challenge, but also had some deficiences we tried to fix for GIVE 2.5. We explain these deficiencies and how we fixed them in GIVE 2.5. We eventually present the preliminary results that show that the system L, like the system NA, achieved a very good result both in objective and in subjective metrics.

Domaines: Interface homme-machine [cs.HC]

Alexandre A. J. Denis. The Loria Instruction Generation System L in GIVE 2.5. 13th European Workshop on Natural Language Generation, Sep 2011, Nancy, France. ⟨inria-00636479⟩